Real-Time Visualization in Non-Isotropic Geometries

Eryk Kopczynski, Dorota Celinska-KopczynskaarXiv 2020 · arXiv

Abstract

Visualizing non-isotropic Thurston geometries, such as Solv and Nil, presents significant computational challenges due to complex geodesics and non-intuitive light paths. This paper introduces a novel primitive-based rendering method that enables real-time navigation and Virtual Reality experiences in these spaces. By utilizing precomputed inverse exponential maps and numerical integration, the authors bridge the gap between abstract mathematical topology and interactive computer graphics.

The Challenge of Non-Isotropic Spaces

In standard Euclidean geometry, light travels in straight lines. In isotropic non-Euclidean spaces like the surface of a sphere or hyperbolic space, geometry remains uniform in all directions. However, non-isotropic geometries behave differently depending on the observer's orientation.

Comparison of tessellations across different Thurston geometries, illustrating the transition from isotropic to non-isotropic spaces.

Visualizing these spaces is difficult for several reasons:

  1. Complex Geodesics: Light rays follow paths that often lack simple closed-form formulas, requiring expensive computation.
  2. Multiple Paths: Points can be reached via multiple "shortest" paths, leading to visual artifacts like caustics.
  3. Computational Cost: Traditional ray-marching is often too slow for the high frame rates required by Virtual Reality.

Methodology: Primitive-Based Rendering

The authors propose a shift from ray-marching to primitive-based rendering, where 3D triangles are projected onto the screen. This approach relies on several key technical innovations:

  • Numerical Integration (RK4): For geometries like Solv, the differential equations of geodesics are solved using the 4th-order Runge-Kutta method.
  • Precomputed 3D Textures: To achieve real-time performance, the inverse exponential map ($\log_M$) is precomputed into 3D textures, allowing the GPU to perform fast interpolation.
  • Discrete Tessellations: To manage the "exponential explosion" of space and avoid floating-point precision issues, the world is represented using combinatorial graphs or honeycombs.
A binary tiling representation used to manage exponential space expansion and maintain numerical precision.
  • Parallel Transport: Camera movement is handled via parallel transport along geodesics to maintain a consistent orientation for the observer.

Key Insights: Navigating the "Weird" Geometries

The research highlights unique properties of specific Thurston geometries:

  • The VR Advantage: Primitive-based rendering is crucial for VR. Unlike ray-marching, it ensures that rays from both eyes converge correctly, maintaining stable depth perception in non-isotropic spaces.
  • Solv's Dimensional "Jumping": In Solv, the shortest path between two points on a plane may involve traveling through a third dimension where space is less stretched.
  • Nil and Impossible Figures: The Nil geometry naturally exhibits properties similar to Penrose stairs; moving in a horizontal loop results in a vertical displacement.
  • Taming Infinity: The binary tiling approach allows for the representation of infinite worlds without the limitations of standard 32-bit or 64-bit floating-point precision.

Results and Performance

The proposed method successfully renders complex scenes in Solv, Nil, and twisted $H^2 \times \mathbb{R}$ at interactive frame rates. Error analysis indicates that RK4 with 100 steps maintains a maximum distance error of less than $10^{-4}$, sufficient for high-quality visualization. These techniques have been integrated into the RogueViz engine and the game HyperRogue, providing the first real-time "native" geodesic views of several Thurston geometries.

Limitations and Future Directions

While effective, the current primitive-based implementation faces challenges in regions with multiple-valued maps (caustics), which can cause objects to stretch or disappear. In highly curved "compact" spaces like the Berger sphere, determining which "version" of a vertex to connect remains difficult. Future work may explore hybrid ray-based methods to handle caustics and the implementation of non-Euclidean shaders for more complex lighting and materials.

Summary updated

Abstract

Non-isotropic geometries are of interest to low-dimensional topologists, physicists and cosmologists. However, they are challenging to comprehend and visualize. We present novel methods of computing real-time native geodesic rendering of non-isotropic geometries. Our methods can be applied not only to visualization, but also are essential for potential applications in machine learning and video games.
BibTeX
@misc{kopczynski2020real,
  author = {Eryk Kopczynski and Dorota Celinska-Kopczynska},
  title = {{Real-Time Visualization in Non-Isotropic Geometries}},
  year = {2020},
  eprint = {2002.09533},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2002.09533},
}
Figure